On Tue, 2015-01-06 at 08:03 +0100, leroy christophe wrote: > Le 05/01/2015 19:12, Joakim Tjernlund a écrit : > > On Mon, 2014-12-22 at 11:14 +0100, Christophe Leroy wrote: > >> On powerpc 8xx, in TLB entries, 0x400 bit is set to 1 for read-only pages > >> and is set to 0 for RW pages. So we should use _PAGE_RO instead of _PAGE_RW > >> > >> Signed-off-by: Christophe Leroy <christophe.le...@c-s.fr> > > Hi Christophe, been meaning to look over all you recent 8xx MMU/TLB patches > > but got so little time :( > > > > This is very cool (not sure if there will be a performance gain) but .. > I think every saved cycle is worth it. > Before I did any modification: > * ITLBMiss was 28 instructions. > * DTLBMiss was 32 instructions. > Now, (No MODULES, no CPU6, no CPU15): > * ITLBMiss is 15 instructions > * DTLBMiss is 24 instructions > >> > >> > >> diff --git a/arch/powerpc/include/asm/pgtable-ppc32.h > >> b/arch/powerpc/include/asm/pgtable-ppc32.h > >> index caf094a..b4e0c3b 100644 > >> --- a/arch/powerpc/include/asm/pgtable-ppc32.h > >> +++ b/arch/powerpc/include/asm/pgtable-ppc32.h > >> @@ -178,9 +178,10 @@ static inline unsigned long pte_update(pte_t *p, > >> andc %1,%0,%5\n\ > >> or %1,%1,%6\n\ > >> /* 0x200 == Extended encoding, bit 22 */ \ > >> - /* Bit 22 has to be 1 if neither _PAGE_USER nor _PAGE_RW are set > >> */ \ > >> + /* Bit 22 has to be 1 when _PAGE_USER is unset and _PAGE_RO is set > >> */ \ > >> rlwimi %1,%1,32-2,0x200\n /* get _PAGE_USER */ \ > >> - rlwinm %3,%1,32-1,0x200\n /* get _PAGE_RW */ \ > >> + rlwinm %3,%1,32-1,0x200\n /* get _PAGE_RO */ \ > >> + xori %3,%3,0x200\n \ > >> or %1,%3,%1\n\ > >> xori %1,%1,0x200\n" > >> " stwcx. %1,0,%4\n\ > > ... here I expected to loose the existing xori insn instead of adding one? > > > > > Well, I could have xored the PAGE_USER bit instead, but in that case, it > is not anymore an 'or' but an 'and' that has to be performed between the > bits, and then all other bits must be set to 1, or the result of the > 'and' shall be inserted using 'rlwimi'. So it would be more > modifications than just adding an xori, and not less instructions.
How about "andc %3,%3,%1; rlwimi %1,%3,0,0x200" instead of the "xori, or, xori" sequence? -Scott -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
